kernel-packages team mailing list archive

[TJ <ubuntu@xxxxxx>]

I did some work upstream but it became too complicated. Others have also
tried but nothing has changed. In the meantime the hardware out there
that has a 32-bit limit imposed by its southbridge on 64-bit CPUs is all
now old legacy gear so I don't foresee the kernel fixing this, ever.

It still remains a problem for the hardware out there still in use where
the user wants to install and use the maximum RAM possible but the
32-bit limitation prevents the PCIOMEM addresses being remapped above
the 4GB boundary.

** Changed in: linux (Ubuntu)
       Status: Incomplete => Triaged

-- 
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to linux in Ubuntu.
https://bugs.launchpad.net/bugs/342926

Title:
  No PCI IOMEM space available below 4GB

Status in The Linux Kernel:
  Confirmed
Status in “linux” package in Ubuntu:
  Triaged
Status in “nvidia-graphics-drivers-180” package in Ubuntu:
  Invalid

Bug description:
  Binary package hint: nvidia-glx-180

  This can affect almost all releases (6.10 7.04, 7.10, 8.04, 8.10,
  9.04+)

  It usually only shows up dramatically with video cards that have large
  memory (e.g. 256MB) and on systems that

  a) have 3GB or more RAM and/or
  b) have 64-bit CPUs on 32-bit north-bridge chipsets (e.g. Intel 945)

  Typical symptom: "failed to load the nvidia kernel module"

  This issue could affect all versions of the Nvidia proprietary drivers
  and others.

  Recently I've seen several users in IRC #ubuntu asking for help but
  without solving the issue. When I saw Keith Dewitt asking the same
  question on 2009-03-14 I arranged with him to access his system via
  SSH and a multiuser screen session to diagnose the issue. Keith was
  very patient and supportive and his assistance led directly to this
  discovery.

  There are also threads on the nvidia forums with the same issues.

  The symptoms are that users report that systems with Nvidia-based
  graphics cards won't start the X server successfully. There are a wide
  range of reports that don't immediately pin-point the cause.

  The fact the nvidia kernel module failed to load is the biggest clue.

  Check dmesg and /var/log/kern.log for something along these lines:

  [   20.137717] nvidia: module license 'NVIDIA' taints kernel.                   
  [   20.412849] ACPI: PCI Interrupt Link [LNEB] enabled at IRQ 18                
  [   20.412858] nvidia 0000:02:00.0: PCI INT A -> Link[LNEB] -> GSI 18 (level, low) -> IRQ 18                                                                    
  [   20.412862] NVRM: This PCI I/O region assigned to your NVIDIA device is invalid:
  [   20.412862] NVRM: BAR1 is 256M @ 0x30000000 (PCI:0002:00.0)                  
  [   20.412865] NVRM: This is a 64-bit BAR mapped above 4GB by the system BIOS or
  [   20.412865] NVRM: Linux kernel. The NVIDIA Linux graphics driver and other   
  [   20.412866] NVRM: system software do not currently support this configuration
  [   20.412867] NVRM: reliably.                                                  
  [   20.412872] nvidia: probe of 0000:02:00.0 failed with error -1               
  [   20.412887] NVRM: The NVIDIA probe routine failed for 1 device(s).           
  [   20.412889] NVRM: None of the NVIDIA graphics adapters were initialized!     

  Also check /var/log/Xorg.0.log or /var/log/Xorg.0.log.old for this
  tell-tale:

  (--) PCI:*(0@2:0:0) nVidia Corporation GeForce 8400 GS rev 161, Mem @ 0xfd000000
  /16777216, 0x130000000/268435456, 0xfa000000/33554432, I/O @ 0x0000ec00/128, BIO
  S @ 0x????????/131072

  (**) NVIDIA(0): Depth 24, (--) framebuffer bpp 32
  (==) NVIDIA(0): RGB weight 888
  (==) NVIDIA(0): Default visual is TrueColor
  (==) NVIDIA(0): Using gamma correction (1.0, 1.0, 1.0)
  (**) NVIDIA(0): Enabling RENDER acceleration
  (II) NVIDIA(0): Support for GLX with the Damage and Composite X extensions is
  (II) NVIDIA(0):     enabled.
  (EE) NVIDIA(0): Failed to load the NVIDIA kernel module!
  (EE) NVIDIA(0):  *** Aborting ***
  (II) UnloadModule: "nvidia"

  Notice that the video card's IOMEM allocation is at 5GB
  (0x130000000/268435456 = 5GB/256MB).

  /var/log/dmesg shows:

  [    0.000000]  BIOS-e820: 0000000100000000 - 0000000130000000
  (usable)

  [    0.477047] pci 0000:00:0b.0: BAR 9: can't allocate resource
  [    0.477047] pci 0000:02:00.0: BAR 1: can't allocate resource 

  lspci -nn reveals:
  00:0b.0 PCI bridge [0604]: nVidia Corporation MCP73 PCI Express bridge [10de:056
  e] (rev a1)
  02:00.0 VGA compatible controller [0300]: nVidia Corporation GeForce 8400 GS [10
  de:0404] (rev a1)

  lspci -vvnn -s 02:00.0

  02:00.0 VGA compatible controller [0300]: nVidia Corporation GeForce 8400 GS [10
  de:0404] (rev a1)                                                               
          Subsystem: eVga.com. Corp. Device [3842:c738]                           
          Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Step
  ping- SERR- FastB2B- DisINTx-                                                   
          Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- 
  <MAbort- >SERR- <PERR- INTx-                                                    
          Latency: 0, Cache Line Size: 64 bytes
          Interrupt: pin A routed to IRQ 18
          Region 0: Memory at fd000000 (32-bit, non-prefetchable) [size=16M]
          Region 1: Memory at 130000000 (64-bit, prefetchable) [size=256M]
          Region 3: Memory at fa000000 (64-bit, non-prefetchable) [size=32M]
          Region 5: I/O ports at ec00 [size=128]
          Expansion ROM at febe0000 [disabled] [size=128K]
          Capabilities: <access denied>
          Kernel modules: nvidia, nvidiafb

  The reason for the failure is that the video chipset's PCI IOMEM RAM
  (in this case 256MB) cannot be allocated in the PCI IOMEM region below
  4GB (from 3GB-4GB) since other devices have already been given
  assignments that mean there isn't a 256MB gap available on a 256MB
  boundary - in other words, either at 3GB or 3.25GB.

  The host system has 4GB of system RAM which causes the kernel to
  prevent the use of the 3GB-3.25GB range, and other allocations prevent
  the use of other ranges.

  Workarounds:

  a) alter the BIOS video IOMEM position to below 4GB (on 64-bit architectures), or
  b) some BIOSes allow setting the "top of memory below 4GB" If so, set it to less than or equal to 2.75GB and try progressively lower values - one should allow a hole large enough for the video IOMEM to fit starting at 3GB (below 4GB), or
  c) reduce the system's RAM to less than 2.5GB to leave sufficient free space.

  For the last year I've had an ongoing project to write a completely
  new PCI IOMEM system for the Linux kernel. I've added the mainline bug
  report that triggered the development to this report. Additionally,
  here is a link to my Wiki describing the issue and solutions.

  http://tjworld.net/wiki/Linux/PCIDynamicResourceAllocationManagement

To manage notifications about this bug go to:
https://bugs.launchpad.net/linux/+bug/342926/+subscriptions